Real time voice identification based gear control system in LMV using MFCC

Speech recognition and speaker recognition have wide range of applications in security systems and smart home designs. In this paper we discuss a method by which text dependent speaker recognition can be used to control gear shifting in light motor vehicles which could be helpful for people who lost one hand in accidents to drive cars. Speaker recognition involves two processes namely feature extraction and feature matching. In feature extraction we extract the dominant features from the voice of the speaker for standard text commands during the training session. There are methods such as Linear Predictive Coding (LPC), Mel Frequency Cepstral Coefficients (MFCC) used for feature extraction. After obtaining these features we form a codebook where characteristics of all the speakers are stored. In feature matching we compare the characteristics of the speaker and intelligent decision making based on the predefined threshold identifies the speaker i.e., driver in our scenario. Hidden Markov Model, Gaussian Mixture Model, Vector Quantization and Neural network as multiclass classifier are some of the methods used for feature matching, while here we make use of neural network. Once the command of the driver is detected, then the gear shifting can be done by an electro mechanical system.